Guillermo Rauch on AI, Scaling Vercel, and The Future of Web Apps

Hey everybody, this is Ben Kaznoka, co-founder and partner at Village Global, a network-driven venture firm. And this is our podcast where we go deep on all things business and technology with world-leading experts. All right, good afternoon. I was thinking we got the little fireplace behind, we could have actually had a fireside chat in front of the actual fireplace, which is a rare opportunity.

It's a gigantic AI or side chat.

Yeah, it would be indeed. Um, well, we have a, a giant of Silicon Valley with us here today in Guillermo, who's atop one of the most important companies in tech today in Vercel. And so we're gonna have a bit of a conversation about the founding story, lessons, AI, what's going on in all things tech, and then, uh, we'll open it up for broader discussion and, uh, and questions. Guillermo, thank you for being here. We were last together, uh, at our other event, The Grove for Unicorn Founders, where you had the tall order of following Jensen Huang. Theatrics, and presented—

He gave us chips, remember?

That's right. Yeah, he gave free H100s to everyone who attended, meaning potato chips that were branded Philip Kohler.

He said, look under your chair, there's free chips. And we looked, and it's like literally a bag of chips with his face on them. Yeah.

So, but great to have you with us. And we have a bunch of early stage founders, as well as some angels and GPs. So let's start with just getting everyone on the same page about the founding story of Vercel, where the company stands today, and maybe importantly, how do you interface with the early-stage startup ecosystem?

Yeah, founding story. I would say there's many checkpoints in time. Once upon a time, I was, uh, I was born and, and grew up in Argentina, and I became obsessed with frontend engineering and open source in the early days. And I always tell people that deploying a hyperlink to the open web was my raft out of Argentina, which is a little bit of an exaggeration because Argentina has a lot of problems, but it wasn't too desperate of a situation, but it allowed me to connect with the world. It got me a job in, through an open source library called Mutools. I got a job internationally. I left Argentina, moved to the Bay Area. So frontend engineering has always been this thing that has tremendous alpha. For companies. If you make a really delightful web experience, if you make a really delightful UI, you're way more likely to capture customers, retain them, turn them into loyal fans. And for certain, uh, industries like e-commerce, you dramatically increase conversion. Having a better front end is better for your business. So I decided to start a company around the idea of maximizing the potential of the front end on the web. And, uh, that's how Vercel was born. So, um, we created a framework called Next.js, which is sort of our open source framework that allows you to create very dynamic and, and very fast web applications. Next.js is very interesting because it grew out of me scratching a niche. Uh, I wanted to create a website for my startup at the time. The founding story is actually, this is a checkpoint that was really, really interesting. So, I sold my first company to WordPress and I left and I knew that I wanted to start another startup and I wanted to use the best practices available at the time. So there were two that were really popular on Twitter and Hacker News, which is how one decides what technologies to use. Of course, one was React.js. So Meta at the time was called the Facebook or Facebook. They've been through three rebrands now. I think about it. That's funny. So they, uh, open sourced what they had discovered to be an incredible way of creating front-end interfaces. Um, and I call it a discovery more so than an invention because it's such an awesome way of structuring very, very complex front-end applications. They open sourced it, they gave React to the world. Kudos to Meta. They've now since open sourced Llama, PyTorch, a lot of amazing things. And then simultaneously Google had open source Kubernetes and me, I'm trying to start a company. I said, I'm gonna use Kubernetes and I'm gonna use React. And then I think it was like 4 weeks later and I maybe had one website up and running and it, it had taken me, an expert that had dedicated their entire life to frontend engineering, weeks just to set up all of the necessary infrastructure and frameworks to actually just publish a website on the internet. And this is like 8 years ago. So that's when the spark happened for me. What if I could make it take seconds instead of weeks? And what if I could give this gift of tremendous iteration velocity and automation to every company in the planet? And it was all predicated on this idea of iterating really fast. Give the developer really good tools. We became known for prioritizing this idea of developer experience or DX. And from there we grew a pretty large cloud infrastructure business that now sits at over $100 million in annualized revenue.

Awesome. And, and talk about how you work with startups and the value prop and the sort of partnership, uh, motions with startups today.

Yeah, what's awesome about Vercel is we're the frontend cloud, so we need the rest of the cloud ecosystem in order to successfully deliver a, a complete web application, right? So an example is when you go to ChatGPT, you're going to a Next.js website that under the hood connects to cloud services, namely of course the ChatGPT API and a bunch of other cloud services to do things like login. For example, it uses Auth0 for login. Every time you, you authenticate, They didn't roll out login from scratch. Same when you go to the, your billing details of ChatGPT, it uses Stripe as an API. So what's amazing about modern engineering, especially at scale, is that a modern cloud-native application is a nexus of all kinds of services that sort of convene at the frontend layer. To deliver the, to the user an awesome experience. What Vercel tries to do is the opposite of shipping the org chart. When you go to an application that's hosted on Vercel, you should have an amazing experience even though there's all these modules under the hood. So the opportunities for startups in our ecosystem is there's so much upside in joining this family of services that are very easy to integrate, We provide APIs so that companies can make one-click install services that they can offer to our developers. Next.js now has over a million monthly active developers. So whenever you create a Next.js-friendly integration or API, you're accessing sort of the, this gigantic pool of developers that are operating at all levels of scale. So I always remind people, when you learn Next.js, You're not just learning a frontend framework skillset, you're learning how to deliver applications at the scale of ChatGPT or Claude or Perplexity, but also the largest e-commerce customers in the world, like nike.com and adidas.com are both powered by Next.js. But also every side hustle that you see on Hacker News, not, maybe not every, but like a lot of them, like if you wanna create a quick experiment, that kind of developer is also using Next.js. So what we've been able to accomplish with the help of the community is a framework that scales up and down really nicely. So if you bet on integrating with it, you get access to this massive distribution.

Okay, let's talk a little bit about AI and then we'll, we'll close with company building entrepreneurship lessons, then we'll open it up for the, for the broader discussion. So on AI, you mentioned you sold your, uh, previous company to Automatic. Yeah. You have a lot of views on open source versus proprietary. You've made the argument that in the long run, open source wins. Yes. In the history of software.

And open platforms.

Open platforms. What's your mental model for thinking about AI today in that context?

Yeah, I created this provocative framework around open versus closed. So when Unix came out, there were a lot of proprietary versions of Unix that nowadays, if I say this, if I said their names in this room, like most people have not even heard heard of them, right? So like there was AIX, there was HP Unix, there was Solaris, there was SCO, SCO, and all of these variants of proprietary Unix ended up just dying and relegated to legacy. And the thing that ended up winning was Linux, which was this open source initiative. By some quote unquote random guy in Finland that literally introduced the project to a mailing list calling it, this'll ne— I think this'll never amount to anything more than a toy. So the power of ecosystems like Linux was predicated on the community that was built and the distributed global infrastructure that was built around it. And the paradox is it actually took longer for Linux to catch up to the proprietary Unix competitors So they had all kinds of advantages, like better hardware support and mainframe integration. Linux always struggled with drivers. Like I could never configure my Wi-Fi on my Linux desktop, but it doesn't matter because this movements of, of this communities are so overwhelmingly difficult to compete with over time. So I try to apply that, that mental model to, to OpenAI in, in the sense of like open source AI. The generic stuff like that, yeah. Where we have to be cognizant that some technical advantages in the short term do not guarantee long-term vitality. So I try to sort of always keep in mind that we have to hedge our bets. And I think if I were to bet over the long haul, my heart and the history background that I've gotten from learning about Linux and other technologies would make me bet on open.

So does that mean that just take OpenAI, the company specifically, you're a seller, not a buyer at $87 billion valuation?

So it's hard to say about valuations and trading stocks. What I can tell you is Vercel's position is developer experience is one of the sort of things that end up deciding how people choose one technology or the other. In fact, if I go back to the origin story of Next.js, React was a very, very exciting technology, but Next.js, the, the way that I explain it is React was like the engine, but companies needed a car. They didn't want to assemble every single piece of the stack every single time. And that's why Vercel ended up being successful. With AI, think about the same. So the LLM is just one specific component. And one of the things that we're, we're building at Vercel is what we call the AI SDK, which is a developer experience that sits on top of all the models. And I'm not gonna say it commoditizes them because all these models have different strengths in different benchmarks. But what I recommend to developers is to actually create that optionality even in their code base. Don't marry yourself only to one model. Buy the index fund of all AI instead of picking one stock. So that would be the answer to like how I think about OpenAI.

In a more macro sense, just as a sidebar, The concerns over safety in open source AI, overrated, underrated? Like, where do you— you don't seem overly concerned about that, right?

Absolutely not.

You think the existential risk, potential for misuse or abuse doesn't concern you?

We're just talking. So I'm from Argentina. Milei is this super interesting character that is—

The president of Argentina.

Yeah, he's captured the attention of a lot of the Silicon Valley leaders. We hosted a dinner. Here in Silicon Valley with President Millet. And one of the things he said that I thought was really thought-provoking, his framework through which he analyzes a lot of the current events is the libertarian framework of let the market decide everything, et cetera, et cetera. And he applied that to AI safety and thought it was very, very interesting. So he said, when Adam Smith wrote about The Wealth of the Nations, he studied this thought experiment of a pin manufacturing factory. And one of the points that he makes in the analysis is that the ability for this factory to maximize its productivity was predicated on a couple factors. One was the specialization of the tasks that the workers could produce. But ultimately, what decides how many pins get manufactured is the law of supply and demand. If the market really, really wants pins, everything will be reconfigured for this factory to produce more pins. And through competition and deregulation, you're gonna see that the most successful nations in the world are the ones that sort of let that supply determine how many of these pins get manufactured. He applied that to the concept of AI safety and thought it was really interesting because a lot of folks are, are worried that if the, uh, power of these models, the reasoning ability of these models continues to increase, they could turn against human beings. But ultimately what decides whether we fund all of this AI labs, whether we decide to fund NVIDIA, whether we buy all of these stocks or we invest in these AI startups will be the demand from customers to want to buy these AI products, right? So just like the demand for pins was determining how many pins were gonna be manufactured, the demand for AI will determine whether we continue to invest in these models. So if AIs are turning against consumers, it's very unlikely that we will continue to fund AI.

So, but isn't, isn't the idea that we may not know they're turning against us until it's too late? Well, like how, what's the feedback loop on, on, on their rebellion against humanity?

I think it's very simple, right? Like we, we, every time there is a model release for a week, everyone freaks out and said we, we, that this happened to us when we launched v0. And every so often, every, we release a new capability and we get the same feedback from people. For those that don't know, v0 is our ChatGPT for web, web developers. You can go in and you type, build me a frontend application. And we specifically called it v0, by the way, because we made the case that it's version zero. We very much need the human to be in the loop with the current model capabilities to give feedback to the model and get into a conversation and prompt and re-prompt and improve. So first of all, we do not believe that AI will take the entire frontend engineer job. However, that for that one week, everyone's freaking out. Oh my God, Vercel is taking our jobs. I still have a few trolls that pursue me in every conversation that I'm having on Twitter and they say that, oh, like conspiracy theories about taking people's jobs. In reality, I very much am in the camp of like job and capability amplification. We'll make developers so much more productive.

I thought you've, I thought you've said elsewhere that you think developers can be in— there are elements of the development community that will be endangered, that there might be commoditization, sort of transformation of UX designers.

Certain elements of the process? Yeah. So I believe two things.

So what type of developers are going to be worse off in a couple of years, or their jobs won't exist as much?

Yeah, so I believe that developers always want to do more. I believe that we should never pigeonhole them, right? We should never say, your skill is only good for creating just the navigation of this website, or your, your skillset is only good for styling this page that was created by somebody else. And I think what I've been noticing, especially with the rise of this role that has now come to get the name of design engineer, engineers are becoming more and more ambitious, assisted by the innovations in these tools. They can do a lot more. So I can design and I can engineer. A lot of the people that we employ at Vercel are experts in just crafting product experiences.

So design engineer, is that a job title inside Vercel? Correct. And, and so what was that person's job title? But like, if you, what was that used to, used to be called? It was like UX meets software.

Well, some of those people have actually self-actualized, right? Yeah. They used to be just designers.

Yeah.

And some of them just used to be engineers. In fact, I, I would call myself a design engineer when I'm, I'm doing my side hustles during the weekends and testing our own products. And I come from the background of frontend engineering, from a very code-centric, API-centric, algorithm-centric. And I taught myself how to design pretty pixels. So going back to your point about, or your question about, I think that there's a lot of what it takes to design pretty pixels that is actually just statistical. What I realized is early on when I taught myself design, that for example, increasing the amount of padding in a containing box created just more visual beauty. Like, it's just easier to read. It's easier on the eyes. And then Apple has this tremendous studies on, like, what is the right Bezier curve of border radius? Like, if you look at the icons in your home screen, they've carefully designed just what is the most aesthetically pleasant border radius. And what happens is designers, like some of the designers that work at Vercel, actually became almost like AIs themselves. Like they just crawl through websites like Dribbble and Twitter and Figma community files, and they're constantly ingesting data of what good user experiences are. And I believe that a lot of that will be preceded because AIs will do a much better job at consuming massive amounts of data of what good design is. And then the design engineer will be able to provide the frontier of that.

Does anyone have design engineers at your company? Is that a title anyone uses? Okay, interesting. So this is a sort of frontier idea.

This frontier role.

Frontier idea.

Frontier job role.

Yeah. And let's just actually pause and see if anyone has questions or comments on the AI theme. I mean, we're gonna switch to company building and like his lessons learned. Entrepreneurship, raising capital. But before I move on from AI, anyone wanna ask questions or comments? Yeah, Shamrat.

I'm, I'm curious, what do you think of the role of these new copilots for engineers? And is there any part of the engineering workflow that is gonna be like fully automated? I believe that there's gonna be a lot of in specific tasks that used to take an enormous amount of time. That'll be taken over by agents. I think when I think about translating an idea from a sketch or a mockup into a working frontend, I think of that job almost as translation. And what happened to translators? I mean, there is a job called translator today, but the vast majority of us that are not doing something that's very much at the cutting edge of like translating a, a, a, famous book. We're just automating 100% of that task. And I believe that for a lot of people building software, that, that idea of producing that front-end code from a design, or even from the idea of the design, what's really interesting is a lot of people don't know what the best way to present a certain type of data is, period. They need an external influence. They need external data. I, I, a funny story from the early days of Next.js is Next.js started finding a lot of like verticals where it was really, really successful early on. One, one that was, that really stood out to me was, um, property list, property listing websites. And, and in a couple of weeks I got calls by Zillow and Trulia and Hilton.com and teams at Airbnb., and it boggled my mind that they were all building the same thing. You, you probably know this because it's become a standard UI, but a pattern that became really, really popular in those websites is you split the screen and on the right-hand side you have Google Maps and pins, and the left-hand side you have properties. Now, if you want to innovate in this regard, and if you're a traditional non-tech native company, you're not gonna go through like the 100 years of like trial and error to find what the right UX pattern for rendering property listings is, you're gonna go with the statistically correct answer. And on top of that, you're gonna layer your branding. So that entire UX research job and design experimentation and all of that does become, I think, just automated away. But again, I think it's just the foundation, and then there's gonna be human creativity layered on top.

So UI and UX developers, designers are at risk.

I think they're not at risk insofar that expertise is not useful. I think if they can—

They have to level up, right?

They can use AI to augment themselves and now go even further. That's why I think we're seeing the success with this design engineering role because people with the help of these tools and these copilots can now go further. Yeah. Front-end engineers can be so much more ambitious these days.

Okay. Other questions or comments on the AI theme? Yep. We can have Lindsay.

I was just curious if you could talk a little bit about your journey going from open source to commercializing, uh, your product and whether or not you think any of the developments in AI is gonna change that for current companies who are trying to follow a similar pattern. Yeah, it's fascinating cuz I recently had, uh, a couple thoughts around— we've, uh, at Vercel, we've spent a tremendous amount of time analyzing and studying and collecting data around what are LLMs best at in terms of code generation. We have extensive benchmarks and evals around different programming languages, frontend libraries. In fact, a funny story from a couple years ago was GPT was obsessed with producing code that recommended an HTTP library called Axios. But the vanguard of web development had already moved on from Axios. In fact, it was extremely disrecommended by Google. And, uh, as you know, Google has a very large, um, developer relations for, uh, workforce that tries to advise folks on how to create really fast websites such that when you go to Google, you get your 10 blue links. Google is very, very incentivized that when you click that blue link, it's really fast. So they invented this metric called the Core Web Vitals. And a lot of what Vercel does is like help you meet that standard so that you get really good SEO and really good prevalence in search results. And one of the things that Google found is that library was just too heavy. It was created for a time when it was necessary, but now it had proliferated across the entire internet. And guess what happened? These LLMs, I mean, they're smart,, but they're also again, very statistical. So they found lots of code that used Axios. So when you went and you asked, hey, can you write me a quick app to do the kind of demo ware that is very, still very, very prevalent in social media? Can you write me a quick app that does this and this? Boom, Axios. So that's when we started realizing that there's a lot of opportunity to sort of augment these LLMs and, and orient them because left to their own devices, they're just gonna do what is like most commonly recommended, right? So I think for us, one of the things that we found is we have to continue to sort of stay very close to this benchmarks and this evals. And that kind of becomes your IP for your company. Like your differentiation is rooted in like what is the, that advantage that you have. Apply to open source, maybe there's a concern that AIs could ossify certain choices in tooling, right? Maybe now there's more Axios prevalence even though it's not the best idea. So when you're thinking about open source, I think you have to now consider, I call this the new network effect because before you had to worry about developers and their choices and their social, like in GitHub and whatever., but now you have to consider the neural networks recommendations of open source. You also have to consider that LLMs are good at generating certain kinds of code. One another thing we found is they strongly favor, and this is unsurprising, local reasoning. So the example that I give people here is LLMs are much better at producing Tailwind CSS. If you're not familiar with Tailwind CSS, when it produces code, It embeds the information about the style of that code in, in the code. Again, the best practice prior to that library was separating code from style. So you had two files. Guess what? LLMs struggle separating in a larger context. Oh, here's all the styles, I'm gonna generate them, and now I'm gonna generate all the HTML. Guess what? They kind of suck at it, but they're really, really good at Tailwind. So, What I think is very novel is that you might have to design in the future open source patterns in the context of, no pun intended, what LLMs are best at. And you're going to be— have to be very, very cognizant of what LLMs are recommending with regards to your open source project. And then the other layer of that is sort of building the business in open source. I can, I can go into that as well.

Yeah, if people are interested in that. Other, other questions on, on sort of AI specifically? Let's go in the back. In the back and then we can go here for sure. We'll start, yeah, start back there. Go for it. Yeah.

Guillermo, hey, big fan of Vercel. Thank you.

Thank you. Can you introduce yourself too?

Yeah, I'm Gautam. I work at Coinbase and I've been an angel investor for a few years now. And my question is sort of like excited to see the v0. I'm curious to hear a little bit more about the long-term vision of where you see the AI SDK going in particular with integrations with LangChain or any third-party integrations?

Yeah.

Yeah. We, we, I talked earlier about we believe that developers should have optionality with regards to model and we created the AI SDK to give you a very, very thin layer. It's a very minimalistic toolkit to focus on developer experience and the ability to choose your model even, uh, and I think the bet is paying off and not just because of downloads and things like that, But just this week, Google announced that you can now go to your Google Chrome settings and enable an experimental flag for a model that runs inside the browser. It's Gemini Nano, and it's, it's an incredible model. It's really fast. It's obviously more cost efficient than going over to the cloud. And you can do amazing things with it. For example, like in the client without going to the internet, for every keystroke you can do inference. And guess what? We're able to implement that model and we released a, uh, an open source implementation of this model today in a matter of minutes without changing our code base. We plugged, uh, Gemini Nano through Chrome AI into the AI SDK. So the bet is really paying off that there's a need for a very lightweight abstraction layer on top of the models that is also very easy for JavaScript developers. So the bet that I made is Python is the language of research scientists and OpenAI researchers and Jupyter Notebook wielders, but the language of application development is TypeScript and JavaScript. So we're betting that the AI SDK will become how we actually bridge the gap between hand-wavy research and frontier models and things like that to actual value for end users. So that's the bet with ASEK. Can we reduce the time to AI for the average product company in the world? Going back to the design engineer, now a design engineer can write a frontier AI application with, I think the example that we put out with the Gemini models, 4 lines of code.

Guillermo, more about your entrepreneurship journey and lessons from entrepreneurship, but just first a little bit of a personal question. I wonder how it informs your worldview. To what extent does being— growing up in Argentina from 0 to 19, how has that shaped the way you think and work? Like, in what ways are you uniquely Argentinian as an entrepreneur, as a person?

I love steak. Yeah. No, actually, uh, actually a lot of people ask me about this. I kind of like it. Um, so number one is just that Like, I love the idea of giving tools to every person in the world regardless of where they are to publish. And I think that there's such a spiritual connection here with, uh, Matt from WordPress who bought my first company because WordPress is all about like letting people publish, uh, in the form of blogs. Vercel is let people publish any kind of application onto the cloud and I think For me, I didn't have the ability to buy software, so open source was, like, amazing. I remember my dad was like, "It's too good to be true. Like, let's go research this Linux thing. It's free? What the hell?" So I think the idea of, like, free access to information and free access to publishing— I really believe that what Apple has done with their, like, approach to getting a developer license getting the DUNS number. Have you ever published to the App Store? You have to get a number called a DUNS. Like, what the hell is that? Like, I just want to publish an app.

So just to get on the Argentinian thing, we were talking at lunch earlier today about people who are up and down versus even keel. The stereotype of people in Argentina, I would say— and I lived in Latin America for a year— is very passionate, very emotional, very enthusiastic. I'm curious, as an entrepreneur, are you like, do you have highs and lows more, or have you had to like develop like an even keel state of mind to endure the journey of entrepreneurship?

It's funny. I do think there's a lot of that lead with your emotions. And it has pluses and minuses. But I believe that For example, Vercel being critical web infrastructure for the world requires a leadership team that has absolutely no biases whatsoever, that has an utmost regard for confidentiality and security. So it, it requires this sort of like incredible, like discipline. And in the way we communicate with the world, we try to remove all of emotion as, as, as much as possible.. But internally, I believe that there is a culture of urgency, for example. I've always been a there's no tomorrow, tomorrow is not promised person. Maybe that's because I'm an immigrant.

How do you cultivate— I think all the startups here want to have a culture like that, a culture of urgency. How do you cultivate that?

I think storytelling is a really powerful thing. I think going back to the sun, so Solaris, that Unix proprietary Unix example that I talked about, the rise and fall of Sun Microsystems is, for the people that were around, is a tragic tale. And I love this thing that Mark Zuckerberg did for the Facebook office where he put the sign of Facebook at the campus, but he used the Sun Microsystems. He kept the back of the sign, was the previous Sun Microsystems logo. And that was just this visceral, always-on reminder of tomorrow's not promised. We could be Sun tomorrow. We could just get acquired or go out of business or start decelerating. Tomorrow's not promised. So I think for me, I feel that always to the bone because in technology, the need for self-actualization is like, there's no other industry that matches it.

So I buy that. Let me just challenge it for a sec though, because I think there is a difference. What's interesting is how do you cultivate urgency among employees for whom this is more a job than it is for you? Like, they're passionate, but they don't own the company.

Well, I try to select missionaries.

Missionaries, yeah.

Not mercenaries.

But I think like, I feel like founders often default to long-term optimistic, and they have to remind themselves—

Short-term pessimistic.

Yeah, the neg— for example, I read—

I heard about this. I think it's Stripe invented that, like short-term pessimist, long-term optimist.

Long-term optimist. That's interesting. Like, I read the memoir about BlackBerry, RIM's rise and fall. And actually, as you were talking, I was thinking it would be interesting for companies to have a book club of, like, all the stories of all these once great companies that failed.

And I collect a lot of those. I actually do this more with the executive team. I remind them, like, do you want to be X? Yeah.

Don't forget. But that is, if you already have the default long-term optimist, optimism, then the kind of fear and anxiety could be good. Like, don't forget, you don't want to be in the next sun. I'm wondering for a broader employee base, is it— would it be better to talk more about like the op— like just, we could be the next Amazon, we could be like, you know, play the upside more in terms of— how do you think about balancing that between your like hardcore exec team and then the long term?

Yeah, I don't think you can get too high on the Kool-Aid of like just pure optimism. I think you do need to infuse it a realism. What I do a lot that is super helpful and constructive is put it through the lens of the customer. So I spend a lot of time talking to customers and I remember powerful anecdotes from the customer and I then contextualize urgency with that storytelling of customers. For example, going back to if I have to explain the importance of change management, for production and mission-critical infrastructure. I use a lot of examples of our healthcare customers and how important it is that every single SLO is met and every single deployment succeeds. Sometimes, and again, this is a day-to-day grind, but sometimes we get the occasional tweet that says, "I just don't know how you guys are so reliable." And it's not that reliability was a magic thing thing that we sparkled onto the system. This quarter we're gonna focus on reliability. I actually shared recently on Twitter, I concluded that reliability is equal parts technology and culture. The culture of reminding yourself of all of those edge cases, of chasing down the edge case. I, I have an internal meme around cosmic rays. So cosmic rays can create this single events of hardware defects. And I believe there was a, a writeup from, uh, S3 where they detected a durability failure on a specific bit flip on a specific hard drive. And that's why they have so much redundancy and they make so many copies and they checksum, they produce cryptographic hashes every single checkpoint of the pipeline because faults can be introduced even randomly. Faults can be introduced by cosmic rays. So what I remind the team is it's so easy for engineers and product people to say, "Oh, it must just have been the internet. Just reboot your computer and it'll sort itself out." And I'm always like, sometimes, you know, PMs have a hard time because if I find a bug and they tell me, "Oh, well, like, just log out and log back in again," I will not do it. I will go, I will have broken software for an entire year Whatever it takes, I wanna see a systemic resolution of that and infuse that culture of, again, the care for reliability and like, it's always on us.

Yeah, I like that. And, and some of us were with Lukas Bywald a couple weeks ago at a Village event, uh, from Weights and Biases, and we were talking about some of his lessons learned and prescriptions in entrepreneurship. And he came back to the simple principle of talk to customers. He's like, every founder knows they should and no one talks to enough customers. Yeah. And it's like this deeply held belief and going deep on the partners.

So I was just in New York for a couple days and maybe this is something I didn't know at the beginning, how important partners were gonna be in fueling the traction of Vercel. And as we get into more complex enterprises, we're almost never alone. So we're gonna be with the partner startups that provide all those services that I spoke about. They were going, there's gonna be the front end and there's gonna be the Auth0s and Stripes and the Stripes of the future, but also there's gonna be the implementation partners that become experts in your technology and build businesses around it. Even the people that build, we have a huge community of people that build and sell templates. And so bringing everyone along for the journey is one of those hacks and talking to them because they're like a multi-customer. They're the customer that creates more customers. And we have that internal meme of like, partners or customers, and you have to be obsessed about interviewing them and bringing those stories.

What are some of your controversial beliefs about entrepreneurship, or what are things you believe that other founders might not?

I believe— I have a conflicting belief, so maybe it's like a cop-out, but I have a conflicting belief. Some of the best investments that I've ever made have been people that are completely obsessed with a problem for year— or an area for years and years and years and years. And that's the entrepreneur that I love to bet on. So I've been obsessed with front engineering for like 20 years. And when I invested in Auth0, I met a team of people that were obsessed with identity. And I remember having that epiphany of like, holy crap, how can you be obsessed with logging in and logging out? It's like, Like, what the hell? And they were telling me all the books that they'd written and the talks that they'd given and like all of the dreams that they had for like better auth. And it was like, home run, you know? And yet, it's not a silver bullet. Like, you always find people that stumble upon ideas. And so that's—

And what's really hard about that is when founders become successful and they tell the story of their startups, the narrative will almost always be, oh, ever since I was 8 years old, I believe in this. And there's a prominent—

Ever since I was in the womb.

Yeah, exactly.

I was passionate. I was obsessed with AI.

Yes, yeah, exactly. There's a prominent example of a very successful tech company today where the CEO has a very touching story about how the mission is so aligned with his deep passions. And I happen to have firsthand knowledge that this was in a venture firm office as an EIR, going through hundreds of different ideas on the whiteboard, analytically evaluate every idea. They arrived at a good business idea, but then on CNN, you're telling a much different story.

Yeah, so maybe the controversial idea, Elon just tweeted about this. He said the CEO of an aerospace company—

Elon just tweeted about this?

Say again?

Elon just tweeted about this?

Yeah, he just tweeted that—

You can just say he.

Yeah, yeah, he.

He just tweeted about it.

Uppercase H, or even lowercase h, tweeted about, well, yeah, duh, The CEO of an aerospace company should know about aircraft and developing aircraft and not about spreadsheets. And it was a critique of Boeing, 'cause Boeing continues to have drama around like—

'Cause Boeing hired like a BCG person or something. Who's the current, the CEO is like some non-engineer.

The suits took over, right? And so maybe the controversial belief is that I think maybe there is a middle where like for certain classes of companies you can pull that off. And then I think for the long-term generational companies, you have to really invest in a leadership team that is intimately familiar with the problem space and lives and breathes that problem. And it might be a meta thing. It might be like just how Amazon became obsessed with customers. And I think there is a sense of like, if there is no obsession, if there is just like opportunism, if there's just like, What is the cool idea that'll get me VC funded? I'm somewhat allergic to that, and I try to stay away from it, but I've also seen it work, so.

Well, and I do think it's like one of these things. We were talking in one of our sessions earlier about how at Village we look for obsession, not just passion, out of founders. Oh, nice. But also, anytime the high status sort of like answer is known by founders, or someone who's interviewing for a job, when the high status answer is known, people can game interviews, game pitches, right? A founder can pitch you and talk about how they're obsessed and say all the right words.

Well, speaking of AI, we're, I was just evaluating our tests that we do for support engineers and—

Like the job interview test?

Yeah. And I realized that like we're getting gamed, that the take-home exercise, it's not because there were too many Delve words, but I could tell that there were some strange similarities between the submissions of different people. And I realize that, you know, like we have to be like even more paranoid now. Not that a cyborg will enter, will walk into your office, but there is a sense of like the verifiability of the story and the verifiability of knowledge that in the age of AI will become—

Well, it's kind of like the commercial world version of what teachers are freaking out with students using AI to generate, do homework, right?

But for entrepreneurs.

Yeah, for entrepreneurs or for job interviews. Yeah. Yeah, interesting. You're famous for valuing, praising the importance of iteration and velocity. And it's something, it's hard to generalize about what characteristics define good entrepreneurs, but speed of iteration seems to be one thing that all great entrepreneurs have in common. How do you promote a culture of iteration and speed of iteration inside Vercel? And what advice do you have for the founders on how to do that in their organization?

Yeah, as we've gone from a single product company to a multi-product company, giving people ownership over the different products and letting them run with that while giving them a framework. For example, we believe a lot, we believe strongly in this idea of analyzing a product experience with fresh eyes and trying to remove all of the assumptions that we have about how much the customer knows about our technology. So, Instead of me going into every product leader and like helping them like QA the product, we're giving them all these principles around like, okay, like look, you are in charge of your product and you can release it, you can tweet about it, you can interview customers, but here's a set of principles that we strongly believe in. For example, around what makes a successful onboarding experience. We, we're obsessed with this zero to one, of any new product.

Uh, yeah, I heard you say you were geeking out over Zapier's, uh, onboarding experience and how they, that you couldn't click the Zapier logo to get out of the onboarding flow. Like they trapped you in that flow. What are other, what are other onboarding, what products have epic onboarding experiences?

So one that got acquired recently, sadly for me, because I wanted them to keep going. Uh, has anyone heard of the X1 credit card or has anyone invested in the X1 credit card? So it just got acquired by Robinhood. The Robinhood credit card is gonna be this company called X1. It was a former product leader at Twitter. I had never experienced such a smooth onboarding experience where I signed up for the credit card. They'd implemented every trick in the book. Not only the, I can't press the logo to interrupt the onboarding, but everything was big. All the buttons were tappable on mobile. I think, I even onboarded myself entirely from a link on a tweet, which if you really geek out on web engineering, one of the things that is really hard is when you find yourself in an in-app browser, it's hellish to do anything. You're more constrained in space. If you open anything in a new window, Apple isolates the cookie jars.. So you're logged in in the in-app web browser potentially, and then you're logged out in Safari. So these people had mastered all of these little details. And I got my credit card into Apple Wallet as a virtual credit card, and then my credit card delivered within 24 hours. Hmm. And just the experience was phenomenal. I was like, that idea of, again, optimizing the mild intent 'Cause a lot of customers on the internet have mild intent. We're all kind of like exploring. We're all doom scrolling a little bit here and there.

Endlessly.

You have to be able to really catalyze. So that was a memorable one for me because it had a hardware component. The credit card showed up. I actually have it here. And I've sent this founder such an overwhelming amount of feedback. He's a saint.

What kind of wallet is that?

Oh, this is a customer of Vercel, Secrid. It's a—

So you can press a button, it pops the card out.

It's the best wallet in the world.

Wow.

So we'll sign around on that. So this is another trick that I advocate for my customers like there's no tomorrow. Like, I will go to war for my customers. And I buy their products, and then by buying their products, I use the website, and then I send them product feedback. This website is phenomenal. But this is the X1 credit card.

It's a— Can I just read that number out loud?

It came out. Yeah. But yeah, so it had the hardware component. So it was like hardware onboarding, virtual onboarding, everything about it was phenomenal, and a beautiful web application as well. So yeah, and again, like simplicity, like the other problem with—

And just finish your point, but we're gonna go to questions in a sec. So come up with your questions for Guillermo, but finish.

Of why you have to narrow the product focus is that there's all of these unknowns. Like maybe you just didn't know that in-app web browsers were gonna ruin your life. So you almost have to have this assumption that like things are gonna get so complex that you have to have a reluctance to extend the product surface. 'Cause what do you think you already tick the box off? And it could be something as simple as my customers can sign up is not quite a given as you might assume it is.

Yeah. Fantastic point.

Questions? Hi, I'm Mila and I'm a former product person, so I have a product I have a question for you. And by the way, love the design engineering job rack. This is great. We have a blog post about it if you want to read it. Love it. So with the Nano ability to run localized model, it's going to enable a lot of personalization on the front end level. Do you see this happening actually? Like when every person in this room, let's say they go to google.com and everybody's going to get their own version. Of google.com? Do you see it only on the web, or do you see it trickling down to the app level as well? And if so, when is your bet? When's it going to happen? Absolutely. Do you want to be Ray Kurzweil and tell us when the singularity is going to happen? Fantastic. Personalized AI. I believe that the entire web will be generative. So you, you go to any product and the experience will be generated for you. We have to move past this idea of, it's like from rendering to generation. It's almost a disruption of React itself. I, I, I got so high on my own supply of React.js and whatever. Like what React invented is like you write this render function and you have all these parameters, but in a world of LLMs, you have like billions of parameters and you have not just what the developer knew were the requirements at the time of implementation. It's like, hmm, maybe, maybe there's 3 states possible. For this UI? Well, no, maybe there's a lot more and they evolve over time. So I think the challenge will be on how we can, uh, retain a very high bar of product quality and predictability because we have to like somehow like tame the craziness of the AI. That's always gonna be the challenge. Function calling, uh, as a tool is amazing because it allows us to sort of build that bridge from the unpredictability of AI and sort of like, and reify it into known schemas of data. So that's one of the ways that you will be able to create UIs that have the generative component, but you also determine these are the, these are the things that I'm interested in for the LLM to generate. Something that's really interesting about the Google example, and I was talking to a lot of the customers of Vercel that are building chat experiences, is that the way to think about it is is we're, we've, most people have only implemented one possible UI of the chat experience, which is the list of messages. What I think is really important, I'm actually bullish on chatbots, is what's really important is that it all starts with the input of the customer. And this was in my opinion, Google's secret sauce is that when you arrive to google.com, all they gave you was the input and they were so ahead of the curve because they started realizing that—

So what do you mean by input there? So you arrive at Google.com, Zapier.com, all they gave you is the input.

The field. Just the field. The input field. Yeah. And there's, going back to like, you can't click the Zapier logo to opt out of their onboarding.

Yeah.

There's almost nothing else to do than to convey your input to them.

Okay, yep.

So what they realized over time is that they started seeing these clusters of queries, right? People started searching for like JetBlue 339. Oh, that's a flight lookup intent. So now I can generate a UI. And in the case of Google, it was incredibly effortful. They needed, uh, tens of thousands of engineers and lots of PMs. And then they were like, hmm, let's spin up the PM for flight lookup. So you're gonna, we're gonna hand you, we're gonna have experimentation infrastructure. We have dynamic rendering, which is what Vercel is trying to commoditize. Like we give you this out of the box and I'm gonna let you focus on that specific segment of queries alone. With AI, now we can sort of automate a lot of that work. We can say from all of these possible inputs and all excluding all of the refusals, 'cause what happened with Amazon's, I dunno if you saw, is like Amazon launched a shopping assistant and something went viral that someone started using it as a Rust copilot. Asking programming language questions on the shopping app on Amazon.com.

The internet remains undefeated as always.

So you have to like determine what your queries are and also what your exclusions are. And based on that, you can create, and going back to like the list of messages, you can now start creating very dynamic user experiences based on the user's intent. And I think that's how a lot of these products are gonna evolve.

Okay, great. No, that's, that's awesome. Let's, Vincenzo, and then— I have a question back to what you were saying. Introduce yourself too. Oh, sorry. Background cybersecurity. I'm running a company that does identity security.

I had a question about what you were saying earlier around the ossification of open source. So how would you counter that?

As in, like, if you had to develop a new product in 3 years or 5 years, say, when LLMs are dominant, Is it to focus on areas where LLMs are particularly bad? So the, the, the cost of switching from the status quo is not that bad, or is it something else?

Like how, how would you approach it? Yeah, you can take the two approaches, right? You can create a, you can ship your product also with the AI tooling so that now people have a, yeah, internally at Vercel at one point we called this DX 2.0. DX 1.0 was us making like the refresh rate of the editor really fast and the compilation time of your Next.js project really fast. v2.0 is to augment that with AI. So you can launch an open source product that also has amazing supplemental AI tooling. That's one approach. The other one is create products that exploit this strengths that the models have and orient your APIs around what the state-of-the-art models are capable of.

Um, how much Bitcoin do you own? Or are you buying or selling Bitcoin, just given the Argentina?

Um, so I have a, I have a really funny story with Bitcoin. Uh, I'm the, I'm the co-organizer of the first ever Dogecoin conference in the world in San Francisco with the co-founder of Doge. We had all the who's who of crypto. This is like 10 years ago.

So I, I was like a Doge, like, like, and that was just meme, like, just fuck it, it's a meme coin?

No, so, so you know how we're talking about like front-end cloud commoditizes services or makes them modular? I believe that crypto will be the true backbone, or at least I consider this to be a case because I'm always paranoid about like what are the potential futures that I'm like, I have to analyze every possible path. What happens if like, which I would love to I would love to see happen if like all of the commerce and transactional infrastructure of the internet becomes a crypto blockchain API. Vercel is still well positioned because you have to create a UI to that experience. And what I believed at the time was like, if I could pay with Doge, I would pay with Doge. So in many ways, Vercel is almost like an opportunity for not just Doge, of course, but like any crypto project to be able to actually kind of give you best of both worlds. You can deliver an amazing customer experience and then plug anything in the back, in through the back. So I've always been super long Bitcoin, Doge, those two actually. And then Solana is really interesting.

At 60+ now, you're holding, buying, selling, what are you doing? Buying.

Buying. Argentina, my dad is like really into stablecoins. He's always into technology, so that's the caveat. But like that, so I remember I went to Argentina a couple years ago and like everybody, like normies, uh, like my high school friends, the popular kids, like all like have crypto wallets and knew about stablecoins and all that, all that stuff.

So it kind of, it kind of like, well, I feel like Argentina is the, yeah, capital of like, if there's a crypto use case anywhere, it's right there.

And, uh, and are using it a lot. Yeah, so I've invested in a number of companies. I'm very, very excited about Lightning. I've been excited for 15 years, so I could be excited like I was excited about the Linux desktop. But I'm very, very excited about providing scalability on top of a rock-solid foundation. I also believe very strongly in the Lindy effect, the fact that Bitcoin has remained undefeated and unhacked and it has the most amazing uptime. Someone as obsessed with cloud infrastructure as I am, the fact that like you cannot compete with that level of reliability is kind of fun to think about, right? Like it's massively distributed, it has its own sort of incentive mechanism to provide availability and to provide nodes. It's self-healing. It's, yeah, cryptographically secure. So it's like an amazing software engineering project.

Well, we at Village, we're always looking for extraordinary clarity of thought in the founders we back, and you really have that in spades, Guillermo. It's really been an honor to host you here. Thank you for the time and the insights, and let's give Guillermo an applause for that. Thank you. Thanks so much for listening to the Village Global Podcast. You can check us out online at villageglobal.vc. We'd love to hear from you, your feedback, your ideas, your inspirations. You can email us at hello@villageglobal.vc.